Design And Development Of Medical Prediction System For Postpartum Hemorrhage

World Health Organization ' s annual World Health Statistics report shows that postpartum hemorrhage(PPH)remains a major cause of maternal mortality worldwide.Therefore,identify risk factors and determine predictive probabilities of those risk factors for primary PPH among women is an extremely urgent and important work.However,due to the discreteness and imbalance problem of PPH data sets,making it more difficult to apply the traditional machine learning method to the risk assessment of PPH.In this thesis,an improved KNN algorithm for PPH data sets,named Adaptive KNN based on Influence Factors(AKNN-IF),is proposed.Traditional KNN algorithm does not consider the degree of influence of feature attributes on label attributes when converting discrete conventional feature attributes into numerical ones.In AKNN-IF algorithm,the influence factors are proposed to solve this problem.And traditional KNN algorithm also needs to perform class equalization on the data set,while AKNN-IF algorithm uses this kind of gender imbalance to calculate each feature class.The proportion of positive samples in the sex value is used to get the influence degree of each feature attribute value on the label attribute as the influence factor of each feature attribute value.Since the causes of PPH are independent,the AKNN-IF algorithm was used to modeling based on different causes respectively.And the experiments results show that the modified KNN classification algorithm is superior to the traditional KNN algorithm.More specifically,the average correct rate,F value and AUC increased by 6.0%,17.0% and13.6%,respectively.Which demonstrates that AKNN-IF can more accurately predict the possible causes of PPH in the early stage.Furthermore,a user-friendly C/S architecture system is development,which use the AKNN-IF algorithm as the core idea.This system implements five functional modules,the login module,instruction module,patient information query module,PPH prediction module and PPH related information module.